基于GA-SVR模型的短期交通流量预测方法研究

发布时间：2019-01-04 16:25

【摘要】：为了提高短期交通流预测精度,寻求最优交通流分组策略,通过对短期历史交通流量数据的分析,运用遗传算法优化支持向量回归机的惩罚参数、核函数参数和不敏感损失函数3个参数,构建了GA-SVR模型。首先对采集的数据采用算术平均值进行了降噪处理,然后根据交通数据特征分为连续5个星期五时间、相邻前5个工作日和当天3种时间周期序列,通过不同时间周期序列确定了最优的训练样本集。最后结合采集的数据进行了验证,并且与传统SVR模型进行了精度对比。结果表明:GA-SVR模型预测精度优于传统SVR模型,且基于当天数据构建的训练样本集总体预测精度最高。
[Abstract]:In order to improve the accuracy of short-term traffic flow prediction and to find the optimal traffic flow grouping strategy, genetic algorithm is used to optimize the penalty parameters of support vector regression machine through the analysis of short-term historical traffic flow data. GA-SVR model is constructed by kernel function parameter and insensitive loss function. Firstly, the arithmetic average is used to reduce the noise of the collected data, and then according to the characteristics of traffic data, the data are divided into five consecutive Friday times, the first five working days and three periods of the same day. The optimal training sample set is determined by different time series. Finally, the collected data are verified and compared with the traditional SVR model. The results show that the prediction accuracy of the GA-SVR model is better than that of the traditional SVR model, and the training sample set constructed on the same day has the highest prediction accuracy.
【作者单位】：中国石油大学(华东)地球科学与技术学院;青岛市黄岛区国土资源和房屋管理局;
【基金】：中央高校基本科研业务费专项资金项目(11CX04012A) 山东省自然科学基金项目(ZR2015DL004) 江苏省海洋动力与声学重点实验室开放基金项目(KHYS1402) 研究生创新工程资助项目(YCXJ2016008)
【分类号】：U491.14

【相似文献】